Say we're interested in the distance between the diagonals $u=(0,0,0)+(1,0,1)t$ and $v=(0,0,1)+(1,1,0)s$ of a unit cube.
The standard formula for the distance between two skew lines $$d=|\mathbf n\cdot (\mathbf a - \mathbf b) |$$ gives $\frac{1}{\sqrt{3}}$. That's likely the answer I am seeking. But, since my original, intuitve approach conflicts with this, I don't want to give up on it.
I considered the distance from the center of a face to the center of an adjacent face. That distance is $\frac{\sqrt{2}}{2}$. The distance from a cube vertex to the center of an "adjacent" face is also $\frac{\sqrt{2}}{2}$ (for example, from $(1,0,1)$ to $(\frac{1}{2},0,1)$ or from $(0,0,1)$ to $(\frac{1}{2},0,\frac{1}{2})$). Since the distance between the two lines is unique, I know then that the distance I want is less than that.
Where is that unique segment, perpedicular to both diagonals?
Edited to get the picture right.
I don't have much to add to André Nicolas' answer, except a picture I quickly drew with Mathematica.
With a little imagination one can see that the minimal line indeed makes right angles with either diagonal.
The Mathematica code used to generate the picture is